Open-source AI Security Operations Center: alert fusion, LLM-agent triage, MITRE ATT&CK investigation, and a replayable decision ledger for every agent step. Self-hostable, runs with no API keys, MIT licensed. Ships an MCP server for Claude, Cursor and Continue.
AiSOC v8 — progress tracker
Last updated: 2026-09-29
Current release: v13.0.0 (2026-09-29) · Next: unscheduled.
Packaging is no longer named against a version: it slipped v8.0 to v8.1 to
v8.2 for the same reason each time, and the blocker is registry credentials
rather than code. Each major since v10.0 has been a major for the same kind of
reason — upgrading requires an operator action — not because the v8 line is
superseded. See the per-version sections of CHANGELOG.md for what each one
closed and what it knowingly left open.
This tracker is the at-a-glance view of what has landed across the v8 line and what is still open. It is deliberately short and dated. When it disagrees with the tree, the tree wins — and the tracker is the thing to fix.
A note on this file's history. It sat unchanged from 2026-06-27 through the v8.0 release, describing four wave-1 tasks as "in flight" months after they merged and naming a loose end that had already landed.
README.mdlinked here for "v8.1 packaging work in flight" and a reader found a June snapshot with no packaging in it. A tracker that is not updated is worse than no tracker, because it is read as current.Its root-level predecessor
AISOC_V8_PROGRESS.mdis the sharper version of the same lesson, and the reason issue #362 was closed rather than refreshed. That issue existed so the wave-2 backlog would be readable "without needing to readAISOC_V8_PROGRESS.md", and pointed at the file anyway — a link that has returned 404 since 2026-06-30, when #368 deleted the root tracker and created this one in the same commit. Every item #362 listed is now recorded here (wave-2, in the table below) or inROADMAP.md(wave-3). Re-auditing before closing it was worth doing rather than assuming: one gap it named — Lacework having no config-snapshot writer — was recorded in neither file and is migrated into the T1.2 row below, and the T3.6 row was itself wrong, still listing a durable approval store that had already landed. A tracking issue for a release four majors old is a tracker nobody reads; this file is the one to keep current.
v8.0 — shipped 2026-09-22
Released as close the loop: connecting capabilities the codebase already contained but never called. The recurring finding, which repeated a dozen times, was the mechanism exists, is unit-tested, and has no caller on the path that needs it — a passing test on an uncalled function is indistinguishable from a working feature until someone traces the call graph.
Full inventory: [8.0.0] in CHANGELOG.md.
Post-v8.0 — the credibility floor and four pillars
Landed on main between the v8.0 tag and the v8.1 cut
(#696,
#697,
#707,
#713,
#715,
#717).
The floor came first because four docs described controls that did not exist.
Each was corrected and then implemented, so the claim could return
honestly. Then the four pillars: context graph (Neo4j schema v1.1 plus real
migration runners for all three non-Postgres stores), recursive investigation
(run_with_tools had zero production callers), the per-capability action
contract, and a SOC-agent benchmark a third party can run against their own
agent.
Claim-to-gate matrix at the v8.0 cut: 108 rows — 99 GATED / 9 PARTIAL /
0 NO GATE. That is a record of where v8.0 landed, not the current tally; the
matrix has grown since. Recount with
python3 scripts/check_claim_gate_matrix.py; do not quote a remembered
figure.
Of the twelve hardening phases, Phase 4 is the only one unchecked, and deliberately: what remains is a funded provider key for the live-agent eval, not code.
v8.1 — shipped 2026-09-23
Wave-2 features (issue #362)
Each item is audited against the tree before any code is written, because the v8.0 lesson is that a capability is often already present and merely unwired. Status is filled in from that audit rather than from the original ticket.
The audit found the backlog wrong in both directions: two items were already built, and four had the capability present with the path that feeds it broken. That is the same shape v8.0 found a dozen times — the mechanism exists, is tested, and has no caller on the path that needs it — so auditing against the tree before writing code is now the first step of a wave rather than an optional one.
| T-ID | Item | What the audit found | Done |
|---|---|---|---|
| T1.2 | Versioned config-snapshot writers, Neo4j :CONFIGURED_AS {ts} |
The Neo4j writer is fully built. The Go provider called a route nobody serves, and could not have called the real one (it has no vault), so snapshots silently never ran while reporting themselves enabled. is_current / valid_from / valid_to were declared and never written, so the documented O(1) lookup matched zero edges. |
partial — Lacework has no get_resource_config, so three of the four providers the ticket named are covered. Five connectors implement it (AWS Security Hub, Azure Defender, GCP SCC, GitHub, Okta); lacework.py is registered and does not. AISOC_SNAPSHOT_ENABLED is also off by default and no compose file sets it, so this runs on operator opt-in only. |
| T2.3 | LLMInputContract across every sub-agent |
Real and fail-closed in services/agents, covering 15 of 16 call sites. services/api had no contract at all across seven endpoints, and the module described as living there did not exist. The no-bypass gate could not see a raw-HTTP LLM call. |
yes |
| T3.2 | Effective-permissions resolvers (Azure / GCP / Okta / GWS) | Already shipped — all five exist, are registered and report coverage: "full". The gap is the snapshot, not the resolver: no connector answers __posture_snapshot__, so four of five return 412. Docstrings still called them scaffolds. |
audited; gap gated |
| T3.3 | Attack-chain ranking + timeline UI | Grouping (fusion, Redis) and weighted ranking (API, Postgres) both exist as independent implementations that never exchange data. AttackChainPanel and AttackStory.tsx are honest about empty states; InvestigationTimeline.tsx rendered a fabricated investigation with no demo gate. |
partial — fabrication gated, the two implementations remain separate |
| T3.5 | Business-context rule engine | Already shipped, and the v8.0 Postgres fix is real. But the two evaluators disagreed on not, so one console-accepted rule silently discarded that tenant's entire rule set at triage, suppressions included. |
yes |
| T3.6 | ChatOps coverage | Every piece existed; three had no caller. Slack Block Kit approvals worked only as a reply to the analyst's own slash command, the Teams card factory had zero senders, and the signed email link pointed at a route that did not exist. Worse: approvals authorized nobody. | partial — authorization, the email route and the durable approval store are all done; the Teams proactive card push remains. The durable store was recorded here as outstanding after it had landed: PostgresTimerStore is wired at services/slack-bot/app/main.py behind migration 062_approval_timers.sql. The genuine remainder is narrower than "Teams" — approval_card() in services/teams-bot/app/cards.py has no caller outside its own tests, and services/teams-bot/app/main.py exposes only inbound routes, so there is no outbound path for an Adaptive Card. The proactive Teams path that does exist (services/actions/app/executors/chatops.py) sends a legacy Connector Card. |
Release integrity
| Item | Status |
|---|---|
| Changelog backfill for the post-v8.0 wave (#682–#719) | done |
RELEASES.md brought up to v8.0.0 (it still announced v7.6.0) |
done |
ROADMAP.md — "v8.0 — Planned" section, stale matrix count |
done |
| This tracker | done |
| Codespaces quickstart (#716) | done |
| arm64 service images (Apple Silicon could not pull any of them) | done |
Screencast + hero.gif assets |
moved to v8.2 — recording needs a stack to point a browser at, and the hosted OSS demo is down pending a billing action |
v8.1.1 — shipped 2026-09-23
An adoption audit with no new capability, run against the whole repository:
installability, architecture comprehension, data provenance, pipeline
connectivity, and whether documented commands work. The headline finding is
that ./install.sh started a compose file with no ingest service, no fusion
service and Kafka disabled — so the most-followed path into the project did
not run the project, and the console a reader saw had been populated by a seed
script.
| Item | Status |
|---|---|
install.sh runs make up (CORE) and then the golden pipeline |
done |
tests/e2e/golden_pipeline/ + golden-pipeline.yml, verified to fail when broken |
done |
scripts/doctor.sh / make doctor |
done |
/livez + /readyz on services/ingest, dialling Kafka |
done |
CORE as the default profile; ClickHouse / Neo4j / Qdrant / OpenSearch / enrichment / connectors to full |
done |
Five fabricated-data surfaces gated behind demo mode; seed_demo.py refuses outside development |
done |
054_alert_provenance.sql — is_synthetic on alerts |
done |
scripts/project_stats.py --check — README figures derived from the tree |
done |
docs/audit/REPOSITORY_REALITY.md |
done |
docs/architecture/README.md rewritten around one event's journey |
done |
docs/testing/CLEAN_INSTALL.md |
done |
| README rewritten for adoption | done |
Consolidated Makefile |
done |
| The three packages that could not be built at the v8.1.0 tag | done |
| CrowdStrike has no normalizer profile and uses the generic one | withdrawn — the wrong diagnosis. The connector does take the canonical path; the real faults were an unread behaviors[].user_name and a canonical field map that recognised only actor. Both fixed in v9.0 — see docs/audit/REPOSITORY_REALITY.md |
OpenSearch is started by full and read by nothing |
withdrawn — checked against services/api, which holds no OpenSearch client, and never against services/threatintel, which uses one unconditionally. Retracted in docs/audit/REPOSITORY_REALITY.md under "Correction (v9.0)" |
Not in v8.1 either: packaging
release.yml already builds, packs and would upload all eight packages on
every tag. The repository's only secret is FLY_API_TOKEN — there is no
NPM_TOKEN and no PyPI trusted publisher — so the upload steps skip with a
warning by design rather than reddening the release.
This is an account action, not an engineering task. Naming it against a release
that cannot perform it is the kind of claim the claim-to-gate matrix exists to
prevent, so packaging is named against v8.2 and becomes a re-tag once the
credentials exist. Registry name state at the v8.1 cut: npm aisoc,
@aisoc/sdk and @aisoc/mcp are unclaimed; on PyPI aisoc is taken by an
unrelated project, so the Python side ships as aisoc-sandbox, and
aisoc-sandbox, aisoc-cli, aisoc-sdk, aisoc-plugin-sdk and
aisoc-detections are all free. One-time setup:
docs/operations/publishing.md.
Deferred by decision — both taken up in v9.0
- Mobile responder console. Recorded here as "not started; no React Native
code in the tree", which was true about React Native and misleading about the
product: the responder console already existed as a PWA under
apps/web/src/app/(responder)/, complete with a service worker, an offline approval queue and Web Push.apps/mobileshipped in v9.0 as a distribution channel on top of it. Its unit tests and type-check run in CI; no device build has been performed. - Plugin marketplace v3 (commercial plugins, revenue sharing). Taken up in
v9.0. The signing and OCI foundations were real; publisher identity was a
stub returning
None, so a submitted plugin was never rejected for a bad signature, and there was no commerce code of any kind in the repository.
Both were labelled "deferred to v8.0" until that label went stale at the v8.0 tag.